Backup Header Below

LALAL.AI Presents Phoenix, a Supreme Solution for Voice Isolation

LALAL.AI, an AI-powered music source separating service, introduces a next-generation AI solution — Phoenix. The neural network is developed to make the vocal extraction process two times faster and clearer than ever before.

Unlike all existing neural networks, Phoenix, in addition to focusing on amplitude processing, doesn’t ignore the phase aspects. Due to architectural improvements, Phoenix analyses more data, allowing for better recognition of the sought-for source’s characteristics and the instruments making up a composition.

“Phoenix processes and splits files into stems twice as fast, delivers higher quality vocal extraction, handles backing vocals much more carefully, and produces significantly fewer artifacts,” says Nikolay Pogorskiy, Lead Engineer. “We established a mathematical foundation and found metrics that let us assess the separation quality.”

The LALAL.AI team has harnessed large data volumes to evaluate the separation quality. It resulted in Phoenix having SDR, a median signal to distortion ratio of about 9.5dB, whereas the previous AI solution Cassiopeia has 8.9dB.

For users, it means significantly more accurate backing vocals separation, more precise main vocal part extraction and much fewer plastic-sounding artifacts in the vocal stems.

As per the LALAL.AI engineers, the methodology leveraged to train Phoenix is about to be applied for the separation of other stems, including bass, drums, piano, synthesizer, electric and acoustic guitars.

The Phoenix AI solution is now available to test on the official LALAL.AI website. Users can split up to 10 minutes for free and compare it to the previous Cassiopeia network. For more details about Phoenix and how it has been developed, please visit this blog post.

Source :

    Other Press Releases